Supervised color constancy using a color chart

One of the key problems in machine vision is color constancy: the ability to match object colors in images taken under different colors of illumination. This is a difficult problem because the apparent color will depend upon the spectral reflectance function of the object and the spectral distribution function of the incident light, both of which are generally unknown. Methods to solve this problem use a small number of basis functions to represent the two functions, and some sort of reference knowledge to allow the calculation of the coefficients. Most methods have the weakness that the reference property may not actually hold for all images, or will have too little information to recover enough of the functions to make an accurate determination of what the color should be. We have developed a method for color constancy that uses a color chart of known spectral characteristics to give stronger reference criteria, and with a large number of colors to give enough information to calculate the illuminant to the desired degree of accuracy. We call this approach "supervised color constancy" since the process is supervised by a picture of a known color chart. We present here two methods for computing supervised color constancy, one using least squares estimation, the other using a neural network. We show experimental results for the supervised calculation of the spectral power distribution of an unknown illuminant. Once this has been calculated, the color of any object with known reflectance can be reliably predicted. We arc developing an extension to allow the prediction of color appearance for an object whose spectral reflectance function is not known. We also propose a method of "incremental color constancy" which determines object color by repeated application of supervised color constancy under changing illumination.